S1 Fig. Brown bear occurrence data and location of the study area in Europe.
https://doi.org/10.1371/journal.pone.0209972.s001
S2 Fig.
Evaluation metrics for 130 candidate models containing different levels of complexity defined by a range of five feature type combinations including linear (L), quadratic (Q), product (P), threshold (T) and hinge (H) features, each evaluated over a range of regularization multipliers ranging from 0 to 10, for (a) the coarse and (b) fine scales of the distribution of the Cantabrian brown bear in Asturias. Evaluation metrics include delta AICc, which is the difference in AICc (Akaikes Information Criterion corrected for small sample sizes, calculated as the sum of the log transformed raw output penalized by the number of model parameters), AUC test, which is the AUC (area Under the receiving operator characteristics Curve) score for the testing data set, AUC diff, which is the difference in AUC scores between the training and testing data sets, and OR min, which is a threshold dependent statistic corresponds to the proportion of testing localities that have MaxEnt output values lower than the value associated with the training locality with the lowest value.
https://doi.org/10.1371/journal.pone.0209972.s002
S3 Fig.
Jacknife evaluations of variable contributions to the (a) coarse and (b) fine scale models. The variables with the highest gain when used in isolation are slope for the coarse scale (a) and forest cover foir the fine scale model (b). These variables therefore seem to have provided the most useful information by themselves for each scale. The variables that decreased the gain most when omitted, and thus possessed the greatest amount of information not present in the other variables, were slope for the coarse scale (a) and population density for the fine scale model (b).
https://doi.org/10.1371/journal.pone.0209972.s003
S4 Fig. Output of the coarse scale model with a 5 x 5 km resolution.
The map presents a clog-log transformation of the raw MaxEnt output, which can be interpreted as a probability of brown bear range occurrence.
https://doi.org/10.1371/journal.pone.0209972.s004
S5 Fig.
Schematic examples of incremental range expansion (a) out of an initial core area as well as (b) a patchy range expansion were no area is occupied two consecutive years, their nestedness values as well as the association matrices used to calculate nestedness.
https://doi.org/10.1371/journal.pone.0209972.s005
S6 Fig. Associations between predicted suitability estimated from the coarse scale model each of the included environmental predictors.
https://doi.org/10.1371/journal.pone.0209972.s006
S7 Fig. Associations between predicted suitability estimated from the fine scale model each of the included environmental predictors.
https://doi.org/10.1371/journal.pone.0209972.s007
S1 Table. Description, source and original format of the 25 environmental variables initially developed for the construction of the models.
Variables marked with * are the ones not correlated and ultimately used in the modelling.
https://doi.org/10.1371/journal.pone.0209972.s008
S2 Table. Variable contribution to the construction of the coarse and fine scale models.
https://doi.org/10.1371/journal.pone.0209972.s009
S3 Table. Centre coordinates of the 5 x 5 km grids classed as bear home range used as bear occurrence data in the coarse scale model.
https://doi.org/10.1371/journal.pone.0209972.s010
S4 Table. Centre coordinates of the 1 x 1 km grids that contained a bear observation used as bear occurrence data in the fine scale model.
https://doi.org/10.1371/journal.pone.0209972.s011